@zfigura I asked about this a bit at WineConf, I understand it should be alright for now to have a dedicated fd for user alerts, but it would be nice to investigate other means to notify waiting threads, using signals for instance.
That's frustrating to hear, because I actually initially designed the kernel pathces that way, explicitly asked Alexandre about that early on in development, and got the feedback that it would be better to make it an fd just like all the others. Not to mention that using signals would require some ppoll-like blocking logic which would have made the ntsync driver no longer fully self-contained.
So that I don't go down a wrong path again, Alexandre, can you please explain what's wrong with the current approach? Why would signals be better?